Method

Four movements, no shortcut.

No execution before validation. No residual dependency after delivery. The method is transparent, the design is documented, the handover is complete.

01.

First conversation, no stopwatch.

No quote without listening. I take the time needed for "yes I want AI" to become "let's talk about your real needs". This is also when the risks you haven't seen surface. Some projects end here, because the right tool doesn't yet exist, or because there's nothing to build.

02.

Architecture on paper.

I hand you a design document before any execution. Every choice is documented and justified. Open decision points are exposed, not hidden. You understand what will be built, why, and what it implies in terms of dependencies and hosting.

03.

Execution on validation.

Never before. Build starts when the design is signed off. No surprises, no silent technical drift. Mid‑course changes are discussed, priced when needed, and tracked.

04.

Full handover.

GitHub repository, rationale guide, twelve months of minor changes included. No subscription. No disguised maintenance contract. You walk away with the tool, the access credentials, and the understanding of its limits.

Engagements

Upstream to design, downstream to put things back on track.

I., Upstream

Prevention.

You come early, to build cleanly. Scoping a project that's starting, choosing an architecture, reviewing critical prompts, preparing a launch. The earlier I'm called in, the less there is to correct.

II., Downstream

Back on track.

You come late, with a tool that has turned against you: a prompt that hallucinates, a site that doesn't convert, a document that creates a legal problem, a deliverable that doesn't hold up. I pinpoint the conceptual flaw. I fix it.

Types of engagement

The right technical answer, not the biggest one.

The tool I recommend depends on the real need. Not on volume, not on buzz.

  • I. Prompt and workflow.Automation via Make, n8n or Zapier for repetitive tasks with predictable format. Quick to set up, minimal dependency.
  • II. Dedicated model or classical algorithm.Sometimes generative AI isn't the right answer. An algorithm, a specialised model, a script will do. Less cost, more reliability.
  • III. Multi‑model architecture.Several models articulated, each on what it does best, with clear role separation and traceability. Only when the previous options aren't enough.
  • IV. Custom connectors.Your tools wired to your databases, your servers, your domain formats. Your server, your data, documented protocols.
  • V. GDPR audit of existing AI tools.For organisations that have already deployed AI tools, agents, prompts in production, informal team usage of cloud LLMs: data flow mapping, identification of compliance gaps, prioritised corrective recommendations.
Commitments

What you won't get.

  • 01 No forced subscription. You own what is delivered. No disguised annual licence.
  • 02 No execution before validation. The architecture document comes before the first line of code.
  • 03 No hidden dependency. External providers are listed, recurring costs estimated, alternatives mentioned.
  • 04 No oversized recommendation. If a workflow is enough, I won't sell you an architecture. If nothing should be built, I say so.
  • 05 No bolted‑on compliance. GDPR analysis is done at architecture, not after the fact. If compliance dictates a technical choice, it is made upstream, not retrofitted via amendment or contractual clause.

No quote, without listening.

The first conversation scopes your request before any decision to build. Thirty minutes, at the pace it takes.

Book a call